lungs-segmentation-unet-densenet-efficientnet-v.ipynb

{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.10","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"gpu","dataSources":[{"sourceId":7773,"sourceType":"datasetVersion","datasetId":4667},{"sourceId":18613,"sourceType":"datasetVersion","datasetId":5839},{"sourceId":5829404,"sourceType":"datasetVersion","datasetId":3350543},{"sourceId":5840694,"sourceType":"datasetVersion","datasetId":3357509},{"sourceId":5865951,"sourceType":"datasetVersion","datasetId":3343827,"isSourceIdPinned":true},{"sourceId":7043037,"sourceType":"datasetVersion","datasetId":1352627},{"sourceId":133564919,"sourceType":"kernelVersion"}],"dockerImageVersionId":30476,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# IMPORT LIBRARIES","metadata":{}},{"cell_type":"code","source":"!pip install -U scipy","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:28:13.912682Z","iopub.execute_input":"2023-11-30T09:28:13.912982Z","iopub.status.idle":"2023-11-30T09:28:38.582344Z","shell.execute_reply.started":"2023-11-30T09:28:13.912954Z","shell.execute_reply":"2023-11-30T09:28:38.581338Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np \nimport pandas as pd \nimport os\nfrom glob import glob\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nimport tensorflow as tf\nfrom sklearn.model_selection import train_test_split\n","metadata":{"_uuid":"46b2e6fb-8735-4a38-943d-e34a585bba93","_cell_guid":"a773e52d-17d8-47e7-8f9c-c07f72159276","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:28:52.919430Z","iopub.execute_input":"2023-11-30T09:28:52.920292Z","iopub.status.idle":"2023-11-30T09:29:00.556998Z","shell.execute_reply.started":"2023-11-30T09:28:52.920258Z","shell.execute_reply":"2023-11-30T09:29:00.555959Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from keras.preprocessing.image import ImageDataGenerator","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:29:05.880821Z","iopub.execute_input":"2023-11-30T09:29:05.881767Z","iopub.status.idle":"2023-11-30T09:29:05.885728Z","shell.execute_reply.started":"2023-11-30T09:29:05.881730Z","shell.execute_reply":"2023-11-30T09:29:05.884810Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from skimage import img_as_float\nfrom skimage import exposure\nfrom skimage.restoration import denoise_tv_chambolle, denoise_bilateral, denoise_wavelet, estimate_sigma\nfrom skimage.filters import threshold_multiotsu,threshold_otsu\nfrom skimage.morphology import erosion, dilation, opening, closing,disk\nfrom skimage.measure import label, regionprops,find_contours\nfrom skimage.morphology import erosion, dilation, opening, closing,disk\nimport cv2","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:29:08.973779Z","iopub.execute_input":"2023-11-30T09:29:08.974632Z","iopub.status.idle":"2023-11-30T09:29:09.591038Z","shell.execute_reply.started":"2023-11-30T09:29:08.974596Z","shell.execute_reply":"2023-11-30T09:29:09.590290Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from tensorflow.keras.applications import MobileNet,VGG19,EfficientNetV2S\nfrom keras.layers import GlobalAveragePooling2D, Dense, Dropout, Flatten,BatchNormalization,Input,Conv2D,MaxPooling2D,concatenate,Conv2DTranspose\nfrom keras.models import Sequential\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:29:12.128485Z","iopub.execute_input":"2023-11-30T09:29:12.129202Z","iopub.status.idle":"2023-11-30T09:29:12.135229Z","shell.execute_reply.started":"2023-11-30T09:29:12.129170Z","shell.execute_reply":"2023-11-30T09:29:12.134138Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# LOAD DATA","metadata":{}},{"cell_type":"code","source":"all_xray_df = pd.read_csv('../input/sample/sample_labels.csv')\nall_image_paths = {os.path.basename(x): x for x in \n                   glob(os.path.join('..','input', 'data','images*', '*', '*.png'))}\n\nall_xray_df['path'] = all_xray_df['Image Index'].map(all_image_paths.get) #get image path\nall_xray_df.head()","metadata":{"_uuid":"60390a8e-1c8f-4bab-963f-71e3e3a5e8c7","_cell_guid":"7e310777-203b-4794-ad28-5f2464028fe2","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:30:04.913801Z","iopub.execute_input":"2023-11-30T09:30:04.914493Z","iopub.status.idle":"2023-11-30T09:30:10.091237Z","shell.execute_reply.started":"2023-11-30T09:30:04.914456Z","shell.execute_reply":"2023-11-30T09:30:10.090161Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# VISUALIZATION","metadata":{}},{"cell_type":"code","source":"# 1 image can have multiple class\nlabel_counts = all_xray_df['Finding Labels'].value_counts()\nlabel_counts","metadata":{"_uuid":"be98d497-9cda-4402-938b-adf98e6448cd","_cell_guid":"3e1fe2b3-ad35-4edf-96e4-682e7b215a0b","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:30:14.206708Z","iopub.execute_input":"2023-11-30T09:30:14.207097Z","iopub.status.idle":"2023-11-30T09:30:14.222367Z","shell.execute_reply.started":"2023-11-30T09:30:14.207054Z","shell.execute_reply":"2023-11-30T09:30:14.221340Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Top 15 class\nlabel_counts = all_xray_df['Finding Labels'].value_counts()[1:15]\nfig, ax1 = plt.subplots(1,1,figsize = (12, 8))\nax1.bar(np.arange(len(label_counts))+0.5, label_counts)\nax1.set_xticks(np.arange(len(label_counts))+0.5)\n_ = ax1.set_xticklabels(label_counts.index, rotation = 90)","metadata":{"_uuid":"cac93a68-a60c-49ec-aca6-273b28567005","_cell_guid":"9b0d40ac-22d2-4d35-8d12-4b6250be4885","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:30:18.732494Z","iopub.execute_input":"2023-11-30T09:30:18.733125Z","iopub.status.idle":"2023-11-30T09:30:19.094132Z","shell.execute_reply.started":"2023-11-30T09:30:18.733075Z","shell.execute_reply":"2023-11-30T09:30:19.093129Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"=> imbalance dataset","metadata":{}},{"cell_type":"markdown","source":"# DATA PROCESSING","metadata":{}},{"cell_type":"markdown","source":"## One-hot","metadata":{}},{"cell_type":"code","source":"all_labels = ['Atelectasis', 'Consolidation', 'Infiltration', 'Pneumothorax', 'Edema', 'Emphysema', 'Fibrosis', 'Effusion', 'Pneumonia', 'Pleural_Thickening', \n'Cardiomegaly', 'Nodule', 'Mass', 'Hernia'] # taken from paper\n\n# make one-hot\nfor lb in all_labels:\n    all_xray_df[lb] = all_xray_df['Finding Labels'].map(lambda result: 1.0 if lb in result else 0)\nall_xray_df.head(20) \n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:30:25.123779Z","iopub.execute_input":"2023-11-30T09:30:25.124631Z","iopub.status.idle":"2023-11-30T09:30:25.237844Z","shell.execute_reply.started":"2023-11-30T09:30:25.124591Z","shell.execute_reply":"2023-11-30T09:30:25.236901Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Train test split","metadata":{}},{"cell_type":"code","source":"# lấy 40000 sample có trọng số\nsample_weights = all_xray_df['Finding Labels'].map(lambda x: len(x.split('|')) if len(x)>0 else 0).values + 4e-2\nsample_weights /= sample_weights.sum()\nall_xray_df = all_xray_df.sample(5000, weights=sample_weights)\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:31:24.939379Z","iopub.execute_input":"2023-11-30T09:31:24.939858Z","iopub.status.idle":"2023-11-30T09:31:24.957056Z","shell.execute_reply.started":"2023-11-30T09:31:24.939826Z","shell.execute_reply":"2023-11-30T09:31:24.956246Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df, valid_df = train_test_split(all_xray_df, \n                                   test_size = 0.2, \n                                   stratify = all_xray_df['Finding Labels'].map(lambda x: x[:4]))\nprint('train', train_df.shape[0], 'validation', valid_df.shape[0])","metadata":{"_uuid":"d9902613-6543-4d8e-83ea-601cdeaf01f9","_cell_guid":"a6ca4b78-2dbe-4a68-ab82-0a0c084be866","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:31:29.548861Z","iopub.execute_input":"2023-11-30T09:31:29.549235Z","iopub.status.idle":"2023-11-30T09:31:29.569644Z","shell.execute_reply.started":"2023-11-30T09:31:29.549198Z","shell.execute_reply":"2023-11-30T09:31:29.568463Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df.head()","metadata":{"_uuid":"8b6c6133-3f6c-4eea-914e-e8189bdc3aeb","_cell_guid":"6b3bb186-b09d-4d52-8c2b-037aa410f578","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:31:32.991573Z","iopub.execute_input":"2023-11-30T09:31:32.992460Z","iopub.status.idle":"2023-11-30T09:31:33.023591Z","shell.execute_reply.started":"2023-11-30T09:31:32.992413Z","shell.execute_reply":"2023-11-30T09:31:33.022543Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Data Generator","metadata":{}},{"cell_type":"code","source":"IMAGE_SIZE = 256  #128","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:31:56.024360Z","iopub.execute_input":"2023-11-30T09:31:56.024718Z","iopub.status.idle":"2023-11-30T09:31:56.028847Z","shell.execute_reply.started":"2023-11-30T09:31:56.024689Z","shell.execute_reply":"2023-11-30T09:31:56.027880Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"strategy = tf.distribute.MirroredStrategy()\nprint('DEVICES AVAILABLE: {}'.format(strategy.num_replicas_in_sync))","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:31:58.956987Z","iopub.execute_input":"2023-11-30T09:31:58.957373Z","iopub.status.idle":"2023-11-30T09:32:01.605990Z","shell.execute_reply.started":"2023-11-30T09:31:58.957345Z","shell.execute_reply":"2023-11-30T09:32:01.604943Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"AUTOTUNE = tf.data.AUTOTUNE\n\nBATCH_SIZE_PER_REPLICA = 32\n\n#We obtain the BATCH_SIZE dividing by the number of devices. \nBATCH_SIZE = BATCH_SIZE_PER_REPLICA*strategy.num_replicas_in_sync\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:32:15.336970Z","iopub.execute_input":"2023-11-30T09:32:15.337372Z","iopub.status.idle":"2023-11-30T09:32:15.341992Z","shell.execute_reply.started":"2023-11-30T09:32:15.337338Z","shell.execute_reply":"2023-11-30T09:32:15.341150Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"core_idg = ImageDataGenerator(horizontal_flip=True,\n                             )","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:32:18.773935Z","iopub.execute_input":"2023-11-30T09:32:18.774832Z","iopub.status.idle":"2023-11-30T09:32:18.779010Z","shell.execute_reply.started":"2023-11-30T09:32:18.774795Z","shell.execute_reply":"2023-11-30T09:32:18.778031Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_generator = core_idg.flow_from_dataframe(dataframe=train_df,\n                                        directory=None,\n                                        x_col = 'path',\n                                        y_col = all_labels,\n                                        class_mode = 'raw',\n#                                         classes = all_labels,\n                                        target_size = (IMAGE_SIZE,IMAGE_SIZE),\n                                        color_mode = 'grayscale',\n                                        batch_size = BATCH_SIZE,\n                                        shuffle = True)\nvalid_generator = core_idg.flow_from_dataframe(dataframe=valid_df,\n                                        directory=None,\n                                        x_col = 'path',\n                                        y_col = all_labels,\n                                        class_mode = 'raw',\n#                                         classes = all_labels,\n                                        target_size = (IMAGE_SIZE,IMAGE_SIZE),\n                                        color_mode = 'grayscale',\n                                        batch_size = 256)","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:32:26.042704Z","iopub.execute_input":"2023-11-30T09:32:26.043383Z","iopub.status.idle":"2023-11-30T09:32:38.377171Z","shell.execute_reply.started":"2023-11-30T09:32:26.043349Z","shell.execute_reply":"2023-11-30T09:32:38.376327Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.figure(figsize=(8,4))\nplt.xticks(rotation = 90)\nplt.bar(all_labels, train_generator.labels.sum(axis = 0)/train_generator.n * 100)\nplt.title('Percentage of different conditions in train dataset')\nplt.xlabel('Conditions')\nplt.ylabel('Percentage')\nplt.show()\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:32:38.378969Z","iopub.execute_input":"2023-11-30T09:32:38.379476Z","iopub.status.idle":"2023-11-30T09:32:38.688203Z","shell.execute_reply.started":"2023-11-30T09:32:38.379440Z","shell.execute_reply":"2023-11-30T09:32:38.687059Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Convert to TF dataset","metadata":{}},{"cell_type":"code","source":"raw_train_data = tf.data.Dataset.from_generator(\n                lambda: train_generator,\n                output_types = (tf.float32, tf.float32), \n                output_shapes = ([None,IMAGE_SIZE,IMAGE_SIZE,1], [None,len(all_labels)]))\n\n\nraw_valid_data = tf.data.Dataset.from_generator(\n                lambda: valid_generator,\n                output_types=(tf.float32, tf.float32), \n                output_shapes=([None,IMAGE_SIZE,IMAGE_SIZE,1], [None,len(all_labels)]))","metadata":{"_uuid":"4b4a3d14-2a21-46c9-ac28-eb73c3daafa5","_cell_guid":"249c7656-f85f-4510-b390-4809d5dcb0e5","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:32:48.240220Z","iopub.execute_input":"2023-11-30T09:32:48.240981Z","iopub.status.idle":"2023-11-30T09:32:48.312419Z","shell.execute_reply.started":"2023-11-30T09:32:48.240941Z","shell.execute_reply":"2023-11-30T09:32:48.311582Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"raw_train_data","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:32:51.694381Z","iopub.execute_input":"2023-11-30T09:32:51.694841Z","iopub.status.idle":"2023-11-30T09:32:51.700977Z","shell.execute_reply.started":"2023-11-30T09:32:51.694803Z","shell.execute_reply":"2023-11-30T09:32:51.700070Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"one_batch_samples = next(iter(raw_train_data))","metadata":{"_uuid":"d957bf3b-b777-4f3f-8910-25622728666e","_cell_guid":"7cf65806-4e92-4653-9018-c16dfe77b5fc","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:32:54.519541Z","iopub.execute_input":"2023-11-30T09:32:54.520266Z","iopub.status.idle":"2023-11-30T09:32:56.221735Z","shell.execute_reply.started":"2023-11-30T09:32:54.520231Z","shell.execute_reply":"2023-11-30T09:32:56.220931Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"image_samples, label_samples = one_batch_samples","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:32:58.241068Z","iopub.execute_input":"2023-11-30T09:32:58.241782Z","iopub.status.idle":"2023-11-30T09:32:58.245664Z","shell.execute_reply.started":"2023-11-30T09:32:58.241748Z","shell.execute_reply":"2023-11-30T09:32:58.244738Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fig, axes = plt.subplots(4, 4, figsize = (16, 16))\nfor (img, lb, ax) in zip(image_samples, label_samples, axes.flatten()):\n    ax.imshow(img, cmap = 'bone')\n    img_lab_name = np.array(all_labels)[lb.numpy() == 1.0]\n    \n    ax.set_title(', '.join(img_lab_name))\n    ax.axis('off')","metadata":{"_uuid":"f8a3454b-2403-4389-8347-f3f02e518e61","_cell_guid":"0c6da886-03f2-46d9-acff-c3f12d15bf82","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:33:01.048622Z","iopub.execute_input":"2023-11-30T09:33:01.048988Z","iopub.status.idle":"2023-11-30T09:33:03.148465Z","shell.execute_reply.started":"2023-11-30T09:33:01.048959Z","shell.execute_reply":"2023-11-30T09:33:03.147149Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## RESCALE","metadata":{}},{"cell_type":"code","source":"normalization_layer = tf.keras.layers.Rescaling(1./255)\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:17.970553Z","iopub.execute_input":"2023-11-30T09:33:17.970916Z","iopub.status.idle":"2023-11-30T09:33:17.978873Z","shell.execute_reply.started":"2023-11-30T09:33:17.970886Z","shell.execute_reply":"2023-11-30T09:33:17.977826Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"scale_train_data = raw_train_data.map(lambda x,y: (normalization_layer(x),y),num_parallel_calls = AUTOTUNE)\nscale_valid_data = raw_valid_data.map(lambda x,y: (normalization_layer(x),y),num_parallel_calls = AUTOTUNE)","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:21.343772Z","iopub.execute_input":"2023-11-30T09:33:21.344392Z","iopub.status.idle":"2023-11-30T09:33:21.421039Z","shell.execute_reply.started":"2023-11-30T09:33:21.344349Z","shell.execute_reply":"2023-11-30T09:33:21.420270Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## BACKGROUND REMOVAL","metadata":{}},{"cell_type":"code","source":"def background_rm(image):\n    \n    img_list = []\n    for i in range(image.shape[0]):\n        try:\n            img = image[i]\n\n            sumOfCols = np.sum(img, axis=0)\n            sumOfRows = np.sum(img, axis=1)\n\n            for i in range(len(sumOfCols)):\n                if sumOfCols[i] > 0:\n                    x1 = i\n                    break\n\n            for i in range(len(sumOfCols)-1,-1,-1):\n                if sumOfCols[i] > 0:\n                    x2 = i\n                    break\n\n            for i in range(len(sumOfRows)):\n                if sumOfRows[i] > 0:\n                    y1 = i\n                    break\n\n            for i in range(len(sumOfRows)-1,-1,-1):\n                if sumOfRows[i] > 0:\n                    y2 = i\n                    break\n\n            roi = img[y1:y2,x1:x2]\n            roi = tf.keras.layers.Resizing(IMAGE_SIZE,IMAGE_SIZE) (roi)\n            img_list.append(roi)\n        except:\n            img_list.append(img)\n\n    return tf.convert_to_tensor(tf.stack(img_list))\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:26.258613Z","iopub.execute_input":"2023-11-30T09:33:26.259487Z","iopub.status.idle":"2023-11-30T09:33:26.269971Z","shell.execute_reply.started":"2023-11-30T09:33:26.259453Z","shell.execute_reply":"2023-11-30T09:33:26.268931Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def background_rm_tf(image,label):\n    im_shape = image.shape\n    [image,] = tf.py_function(background_rm, [image], [tf.float32])\n    image.set_shape(im_shape)\n    return image, label\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:31.236152Z","iopub.execute_input":"2023-11-30T09:33:31.236490Z","iopub.status.idle":"2023-11-30T09:33:31.241618Z","shell.execute_reply.started":"2023-11-30T09:33:31.236466Z","shell.execute_reply":"2023-11-30T09:33:31.240666Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"rm_background_train_data = scale_train_data.map(background_rm_tf,num_parallel_calls = AUTOTUNE)\nrm_background_valid_data = scale_valid_data.map(background_rm_tf,num_parallel_calls = AUTOTUNE)","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:34.777993Z","iopub.execute_input":"2023-11-30T09:33:34.778377Z","iopub.status.idle":"2023-11-30T09:33:34.859643Z","shell.execute_reply.started":"2023-11-30T09:33:34.778348Z","shell.execute_reply":"2023-11-30T09:33:34.858719Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## REMOVE NOISE","metadata":{}},{"cell_type":"code","source":"def denoise(image):\n    return denoise_tv_chambolle(img_as_float(image), weight=0.01)\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:38.180395Z","iopub.execute_input":"2023-11-30T09:33:38.181317Z","iopub.status.idle":"2023-11-30T09:33:38.185932Z","shell.execute_reply.started":"2023-11-30T09:33:38.181280Z","shell.execute_reply":"2023-11-30T09:33:38.184973Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def denoise_tf(image,label):\n    im_shape = image.shape\n    [image,] = tf.py_function(denoise, [image], [tf.float32])\n    image.set_shape(im_shape)\n    return image, label\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:40.447165Z","iopub.execute_input":"2023-11-30T09:33:40.448072Z","iopub.status.idle":"2023-11-30T09:33:40.452744Z","shell.execute_reply.started":"2023-11-30T09:33:40.448037Z","shell.execute_reply":"2023-11-30T09:33:40.451748Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"rm_noise_train_data = rm_background_train_data.map(denoise_tf,num_parallel_calls = AUTOTUNE)\nrm_noise_valid_data = rm_background_valid_data.map(denoise_tf,num_parallel_calls = AUTOTUNE)","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:43.970263Z","iopub.execute_input":"2023-11-30T09:33:43.970614Z","iopub.status.idle":"2023-11-30T09:33:44.042759Z","shell.execute_reply.started":"2023-11-30T09:33:43.970588Z","shell.execute_reply":"2023-11-30T09:33:44.042005Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## REMOVE THE DIAPHRAGM REGION","metadata":{}},{"cell_type":"code","source":"def rmDiapgRegion(image):\n    img = img_as_float(image)\n    img = np.copy(img)\n    # get Diaphragm Region\n    \n    brightest = img.max()\n    darkest = img.min()\n#     print(brightest)\n    T = darkest + 0.94 * (brightest - darkest)\n    binary = img > T\n\n    closed = closing(binary) \n    dilate = dilation(closed)\n\n    # get high density region\n    l = label(dilate)\n    out = (l==np.bincount(l.ravel())[1:].argmax()+1).astype(np.int64)\n\n    # smoothing region\n#     opened = closing(out)\n    img[out.astype(np.float32) == 1.0] = 0\n        \n    return img\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:48.190181Z","iopub.execute_input":"2023-11-30T09:33:48.190531Z","iopub.status.idle":"2023-11-30T09:33:48.198144Z","shell.execute_reply.started":"2023-11-30T09:33:48.190503Z","shell.execute_reply":"2023-11-30T09:33:48.196949Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def rmDiapgRegion_tf(image,label):\n    im_shape = image.shape\n    [image,] = tf.py_function(rmDiapgRegion, [image], [tf.float32])\n    image.set_shape(im_shape)\n    return image, label\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:50.512357Z","iopub.execute_input":"2023-11-30T09:33:50.512768Z","iopub.status.idle":"2023-11-30T09:33:50.518629Z","shell.execute_reply.started":"2023-11-30T09:33:50.512737Z","shell.execute_reply":"2023-11-30T09:33:50.517363Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"rm_diapg_train_data = rm_noise_train_data.map(rmDiapgRegion_tf,num_parallel_calls =AUTOTUNE)\nrm_diapg_valid_data = rm_noise_valid_data.map(rmDiapgRegion_tf,num_parallel_calls =AUTOTUNE)","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:53.267626Z","iopub.execute_input":"2023-11-30T09:33:53.267985Z","iopub.status.idle":"2023-11-30T09:33:53.337237Z","shell.execute_reply.started":"2023-11-30T09:33:53.267956Z","shell.execute_reply":"2023-11-30T09:33:53.336327Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"rm_diapg_train_data","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:33:55.474710Z","iopub.execute_input":"2023-11-30T09:33:55.475378Z","iopub.status.idle":"2023-11-30T09:33:55.481316Z","shell.execute_reply.started":"2023-11-30T09:33:55.475345Z","shell.execute_reply":"2023-11-30T09:33:55.480398Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# for image,l in rm_diapg_train_data.take(1):\n#     plt.imshow(image[3],cmap = 'bone')","metadata":{"execution":{"iopub.status.busy":"2023-06-14T04:57:50.775033Z","iopub.execute_input":"2023-06-14T04:57:50.775608Z","iopub.status.idle":"2023-06-14T04:57:50.781777Z","shell.execute_reply.started":"2023-06-14T04:57:50.775569Z","shell.execute_reply":"2023-06-14T04:57:50.780753Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## CONTRAST ENRICHMENT\n","metadata":{}},{"cell_type":"code","source":"def contrast_enrich(image):\n    return exposure.equalize_adapthist(img_as_float(image),clip_limit = 0.02)\n        ","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:37:54.568683Z","iopub.execute_input":"2023-11-30T09:37:54.569757Z","iopub.status.idle":"2023-11-30T09:37:54.574194Z","shell.execute_reply.started":"2023-11-30T09:37:54.569716Z","shell.execute_reply":"2023-11-30T09:37:54.573261Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def contrast_enrich_tf(image,label):\n    im_shape = image.shape\n    [image,] = tf.py_function(contrast_enrich, [image,], [tf.float32])\n    image.set_shape(im_shape)\n    return image, label\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:37:57.220206Z","iopub.execute_input":"2023-11-30T09:37:57.220559Z","iopub.status.idle":"2023-11-30T09:37:57.226102Z","shell.execute_reply.started":"2023-11-30T09:37:57.220530Z","shell.execute_reply":"2023-11-30T09:37:57.224923Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"enhance_contrast_train_data = rm_diapg_train_data.map(contrast_enrich_tf, num_parallel_calls = AUTOTUNE)\nenhance_contrast_valid_data = rm_diapg_valid_data.map(contrast_enrich_tf, num_parallel_calls = AUTOTUNE)\n","metadata":{"_uuid":"f6e8f75a-2773-4850-af26-e79c99d2d8dd","_cell_guid":"d09feb0b-b9fe-4bb1-af53-f943bba47fbb","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-11-30T09:37:59.450366Z","iopub.execute_input":"2023-11-30T09:37:59.451115Z","iopub.status.idle":"2023-11-30T09:37:59.521763Z","shell.execute_reply.started":"2023-11-30T09:37:59.451070Z","shell.execute_reply":"2023-11-30T09:37:59.521024Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# input image\nfor image,l in enhance_contrast_train_data.take(1):\n    plt.imshow(image[1],cmap = 'bone')","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:38:01.138677Z","iopub.execute_input":"2023-11-30T09:38:01.139424Z","iopub.status.idle":"2023-11-30T09:38:08.032475Z","shell.execute_reply.started":"2023-11-30T09:38:01.139391Z","shell.execute_reply":"2023-11-30T09:38:08.031478Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## LUNG SEGMENTATION","metadata":{}},{"cell_type":"markdown","source":"A segmentation-based cropped. This function will encourage model to focus on the lungs","metadata":{}},{"cell_type":"code","source":"inputs = Input((IMAGE_SIZE,IMAGE_SIZE,1))\ninput_resize = tf.keras.layers.Resizing(512,512, name = 'resize_input') (inputs)\nconv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(input_resize)\nconv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv1)\npool1 = MaxPooling2D(pool_size=(2, 2))(conv1)\n\nconv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(pool1)\nconv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv2)\npool2 = MaxPooling2D(pool_size=(2, 2))(conv2)\n\nconv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(pool2)\nconv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv3)\npool3 = MaxPooling2D(pool_size=(2, 2))(conv3)\n\nconv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(pool3)\nconv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv4)\npool4 = MaxPooling2D(pool_size=(2, 2))(conv4)\n\nconv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(pool4)\nconv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv5)\n\nup6 = concatenate([Conv2DTranspose(256, (2, 2), strides=(2, 2), padding='same')(conv5), conv4], axis=3)\nconv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(up6)\nconv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv6)\n\nup7 = concatenate([Conv2DTranspose(128, (2, 2), strides=(2, 2), padding='same')(conv6), conv3], axis=3)\nconv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(up7)\nconv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv7)\n\nup8 = concatenate([Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(conv7), conv2], axis=3)\nconv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(up8)\nconv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv8)\n\nup9 = concatenate([Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same')(conv8), conv1], axis=3)\nconv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(up9)\nconv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv9)\n\nconv10 = Conv2D(1, (1, 1), activation='sigmoid')(conv9)\n\nmodel = tf.keras.Model(inputs=[inputs], outputs=[conv10])","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:38:14.502949Z","iopub.execute_input":"2023-11-30T09:38:14.503692Z","iopub.status.idle":"2023-11-30T09:38:14.863184Z","shell.execute_reply.started":"2023-11-30T09:38:14.503654Z","shell.execute_reply":"2023-11-30T09:38:14.862273Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.load_weights(\"/kaggle/input/unet-lung-segmentation-weights-for-chest-x-rays/cxr_reg_weights.best.hdf5\")","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:38:21.636430Z","iopub.execute_input":"2023-11-30T09:38:21.637116Z","iopub.status.idle":"2023-11-30T09:38:21.998954Z","shell.execute_reply.started":"2023-11-30T09:38:21.637072Z","shell.execute_reply":"2023-11-30T09:38:21.998145Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def lung_segment(image):\n    pred = model.predict(image,verbose = 0) > 0.4\n    pred = pred.astype(np.int32) # 512,512\n    img_list = []\n    \n    for i in range(image.shape[0]):\n        l = label(pred[i])\n        z = np.argpartition(np.bincount(l.ravel())[1:],-2)\n\n        out_1 = (l== z[-2] +1).astype(int)\n        out_2 = (l== z[-1] +1).astype(int)\n        combine = out_1 + out_2\n        combine = tf.squeeze(combine)    \n\n        minc_local, maxc_local = 1000,0\n        \n        label_image = label(combine)\n        regions = regionprops(label_image)\n\n        for region in regions:\n            minr, minc, maxr, maxc = region.bbox\n            if minc_local > minc:\n                minc_local = minc\n            if maxc_local < maxc:\n                maxc_local = maxc\n\n#         if maxc_local - minc_local <= 200 or maxc_local - minc_local >= 300:\n#             img_list.append(image[i])\n#             continue\n        img = tf.keras.layers.Resizing(512,512) (image[i])\n        if minc_local - 20 >=0:\n            minc_local -= 20\n        if maxc_local + 20 <= img.shape[0]:\n            maxc_local += 20\n        z = img[:,minc_local:maxc_local]\n        z = tf.keras.layers.Resizing(IMAGE_SIZE,IMAGE_SIZE,interpolation = \"area\") (z)\n        img_list.append(z)\n\n    return tf.convert_to_tensor(tf.stack(img_list))","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:38:28.985040Z","iopub.execute_input":"2023-11-30T09:38:28.985653Z","iopub.status.idle":"2023-11-30T09:38:28.995999Z","shell.execute_reply.started":"2023-11-30T09:38:28.985620Z","shell.execute_reply":"2023-11-30T09:38:28.995044Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def lung_segment_tf(image,label):\n    im_shape = image.shape\n    [image,] = tf.py_function(lung_segment, [image], [tf.float32])\n    image.set_shape(im_shape)\n    return image, label\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:38:31.313341Z","iopub.execute_input":"2023-11-30T09:38:31.313712Z","iopub.status.idle":"2023-11-30T09:38:31.324715Z","shell.execute_reply.started":"2023-11-30T09:38:31.313684Z","shell.execute_reply":"2023-11-30T09:38:31.323117Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"lung_segment_train_data = enhance_contrast_train_data.map(lung_segment_tf, num_parallel_calls = AUTOTUNE)\nlung_segment_valid_data = enhance_contrast_valid_data.map(lung_segment_tf, num_parallel_calls = AUTOTUNE)\n","metadata":{"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def plot_lung_segment_img(image_sample):\n    test_img = tf.expand_dims(image_sample,0)\n    test_img /= 255.\n    test_img = contrast_enrich(test_img)\n    test_img = lung_segment(test_img)[0] # it return batchsize,_,_\n    \n    fig,axes = plt.subplots(1,2,figsize = (15,15))\n    axes[0].imshow(image_sample,cmap = 'bone')\n    axes[0].set_title(\"Original Image\")\n    \n    axes[1].imshow(test_img,cmap = 'bone')\n    axes[1].set_title(\"Segmentation-based cropped\")\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:38:52.962156Z","iopub.execute_input":"2023-11-30T09:38:52.962517Z","iopub.status.idle":"2023-11-30T09:38:52.968922Z","shell.execute_reply.started":"2023-11-30T09:38:52.962489Z","shell.execute_reply":"2023-11-30T09:38:52.967769Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plot_lung_segment_img(image_samples[8])","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:38:56.613153Z","iopub.execute_input":"2023-11-30T09:38:56.613984Z","iopub.status.idle":"2023-11-30T09:39:03.533382Z","shell.execute_reply.started":"2023-11-30T09:38:56.613953Z","shell.execute_reply":"2023-11-30T09:39:03.532424Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# MODEL","metadata":{}},{"cell_type":"code","source":"train_data = scale_train_data.map(lambda x,y: (tf.image.grayscale_to_rgb(x),y)).cache().prefetch(buffer_size = AUTOTUNE)\nvalid_data = scale_valid_data.map(lambda x,y: (tf.image.grayscale_to_rgb(x),y)).cache().prefetch(buffer_size = AUTOTUNE)","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:39:19.389895Z","iopub.execute_input":"2023-11-30T09:39:19.390607Z","iopub.status.idle":"2023-11-30T09:39:19.512972Z","shell.execute_reply.started":"2023-11-30T09:39:19.390572Z","shell.execute_reply":"2023-11-30T09:39:19.512230Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_X, test_Y = next(core_idg.flow_from_dataframe(dataframe=valid_df,\n                                        directory=None,\n                                        x_col = 'path',\n                                        y_col = all_labels,\n                                        class_mode = 'raw',\n#                                         classes = all_labels,\n                                        target_size = (IMAGE_SIZE,IMAGE_SIZE),\n                                        color_mode = 'grayscale',\n                                        batch_size = 1024))\n","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:39:27.179562Z","iopub.execute_input":"2023-11-30T09:39:27.180358Z","iopub.status.idle":"2023-11-30T09:39:50.156873Z","shell.execute_reply.started":"2023-11-30T09:39:27.180323Z","shell.execute_reply":"2023-11-30T09:39:50.155796Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_X_normal = tf.keras.layers.Rescaling(1./255) (test_X)\ntest_X_normal = tf.image.grayscale_to_rgb(tf.convert_to_tensor(test_X_normal))","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:39:57.905268Z","iopub.execute_input":"2023-11-30T09:39:57.905948Z","iopub.status.idle":"2023-11-30T09:39:58.186991Z","shell.execute_reply.started":"2023-11-30T09:39:57.905917Z","shell.execute_reply":"2023-11-30T09:39:58.185941Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_X_process = test_X.copy()\n\ntest_X_process = background_rm(test_X_process)\ntest_X_process = tf.keras.layers.Rescaling(1./255) (test_X_process)\ntest_X_process = denoise(test_X_process)\ntest_X = rmDiapgRegion(test_X)\ntest_X_process = contrast_enrich(test_X_process)\n# test_X = lung_segment(test_X) #use this will degrade performance\ntest_X_process = tf.image.grayscale_to_rgb(tf.convert_to_tensor(test_X_process))","metadata":{"execution":{"iopub.status.busy":"2023-11-30T09:40:00.949549Z","iopub.execute_input":"2023-11-30T09:40:00.949913Z","iopub.status.idle":"2023-11-30T09:41:20.692139Z","shell.execute_reply.started":"2023-11-30T09:40:00.949883Z","shell.execute_reply":"2023-11-30T09:41:20.691111Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## DENSE121","metadata":{}},{"cell_type":"code","source":"input = tf.keras.layers.Input([IMAGE_SIZE, IMAGE_SIZE, 3], dtype = tf.float32)\n\ndense121 = tf.keras.applications.DenseNet121(include_top=False,weights=\"imagenet\",pooling='avg')\ndense121 = dense121(input,)\n\noutput = Dense(len(all_labels), activation = 'sigmoid') (dense121)\n\ndense121_model = tf.keras.Model(inputs=[input], outputs=[output])\ndense121_model.summary()","metadata":{"execution":{"iopub.status.busy":"2023-06-14T05:02:05.03539Z","iopub.execute_input":"2023-06-14T05:02:05.035993Z","iopub.status.idle":"2023-06-14T05:02:09.595479Z","shell.execute_reply.started":"2023-06-14T05:02:05.035959Z","shell.execute_reply":"2023-06-14T05:02:09.59458Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dense121_model.compile(optimizer = tf.keras.optimizers.experimental.Adam(0.0001), loss = 'binary_crossentropy',\n                           metrics = ['binary_accuracy', tf.keras.metrics.AUC()])","metadata":{"execution":{"iopub.status.busy":"2023-06-07T12:54:51.567719Z","iopub.execute_input":"2023-06-07T12:54:51.568091Z","iopub.status.idle":"2023-06-07T12:54:51.602531Z","shell.execute_reply.started":"2023-06-07T12:54:51.568059Z","shell.execute_reply":"2023-06-07T12:54:51.601636Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"steps_per_epoch = train_df.shape[0]  // BATCH_SIZE\nsteps_per_epoch","metadata":{"execution":{"iopub.status.busy":"2023-06-07T12:50:44.220127Z","iopub.execute_input":"2023-06-07T12:50:44.220485Z","iopub.status.idle":"2023-06-07T12:50:44.227043Z","shell.execute_reply.started":"2023-06-07T12:50:44.220454Z","shell.execute_reply":"2023-06-07T12:50:44.22602Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dense121_model.load_weights(\"/kaggle/input/vgg19h5/dense121_normal.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-07T12:50:44.228518Z","iopub.execute_input":"2023-06-07T12:50:44.228884Z","iopub.status.idle":"2023-06-07T12:50:45.025419Z","shell.execute_reply.started":"2023-06-07T12:50:44.228853Z","shell.execute_reply":"2023-06-07T12:50:45.024447Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"dense121_model.fit(train_data, \n            steps_per_epoch = 750,\n#           validation_data = valid_data, \n            epochs = 1, \n            verbose = 1,\n          )","metadata":{"execution":{"iopub.status.busy":"2023-06-07T12:55:00.424033Z","iopub.execute_input":"2023-06-07T12:55:00.424403Z","iopub.status.idle":"2023-06-07T13:04:51.110784Z","shell.execute_reply.started":"2023-06-07T12:55:00.424356Z","shell.execute_reply":"2023-06-07T13:04:51.109774Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# dense121_model.save_weights(\"dense121_normal_.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-07T12:54:43.075782Z","iopub.status.idle":"2023-06-07T12:54:43.076506Z","shell.execute_reply.started":"2023-06-07T12:54:43.076234Z","shell.execute_reply":"2023-06-07T12:54:43.076256Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### NORMAL","metadata":{}},{"cell_type":"code","source":"y_pred_dense121_normal = dense121_model.predict(test_X_normal)\ny_pred_dense121_normal[:5]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-07T12:38:30.590891Z","iopub.execute_input":"2023-06-07T12:38:30.591429Z","iopub.status.idle":"2023-06-07T12:38:34.490424Z","shell.execute_reply.started":"2023-06-07T12:38:30.591398Z","shell.execute_reply":"2023-06-07T12:38:34.489425Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc,f1_score\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_dense121_normal = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_dense121_normal[:,idx])\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_dense121_normal += auc(fpr, tpr)\n\nmean_auc_dense121_normal /= 14\nprint(mean_auc_dense121_normal)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')\n","metadata":{"execution":{"iopub.status.busy":"2023-06-07T12:39:52.945779Z","iopub.execute_input":"2023-06-07T12:39:52.94617Z","iopub.status.idle":"2023-06-07T12:39:53.401523Z","shell.execute_reply.started":"2023-06-07T12:39:52.946139Z","shell.execute_reply":"2023-06-07T12:39:53.400528Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### PREPROCESS","metadata":{}},{"cell_type":"code","source":"dense121_model.load_weights(\"/kaggle/input/vgg19h5/dense121_3.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-14T05:06:36.397049Z","iopub.execute_input":"2023-06-14T05:06:36.397452Z","iopub.status.idle":"2023-06-14T05:06:37.28403Z","shell.execute_reply.started":"2023-06-14T05:06:36.39742Z","shell.execute_reply":"2023-06-14T05:06:37.283052Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred_dense121_preprocess = dense121_model.predict(test_X_process)\ny_pred_dense121_preprocess[:5]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T05:06:37.286108Z","iopub.execute_input":"2023-06-14T05:06:37.286446Z","iopub.status.idle":"2023-06-14T05:06:38.017445Z","shell.execute_reply.started":"2023-06-14T05:06:37.286415Z","shell.execute_reply":"2023-06-14T05:06:38.016469Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc,f1_score\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_dense121_preprocess = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_dense121_preprocess[:,idx])\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_dense121_preprocess += auc(fpr, tpr)\n\nmean_auc_dense121_preprocess /= 14\nprint(mean_auc_dense121_preprocess)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')\n","metadata":{"execution":{"iopub.status.busy":"2023-06-14T05:06:54.793587Z","iopub.execute_input":"2023-06-14T05:06:54.793978Z","iopub.status.idle":"2023-06-14T05:06:55.365556Z","shell.execute_reply.started":"2023-06-14T05:06:54.793947Z","shell.execute_reply":"2023-06-14T05:06:55.364596Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## VGG19","metadata":{}},{"cell_type":"code","source":"input = tf.keras.layers.Input([IMAGE_SIZE, IMAGE_SIZE, 3], dtype = tf.float32)\nprocessed_input = tf.keras.applications.vgg19.preprocess_input(input)\n\nvgg19 = VGG19(include_top = False,weights = 'imagenet') \nvgg19 = vgg19(input)\n\nnormal = BatchNormalization() (vgg19)\nflatten = GlobalAveragePooling2D() (normal)\n\ndense1 = Dense(256,activation = 'relu') (flatten)\n\noutput = Dense(len(all_labels), activation = 'sigmoid') (dense1)\n\nvgg19_model = tf.keras.Model(inputs=[input], outputs=[output])\nvgg19_model.summary()","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:21:19.502532Z","iopub.execute_input":"2023-06-05T03:21:19.503524Z","iopub.status.idle":"2023-06-05T03:21:20.689733Z","shell.execute_reply.started":"2023-06-05T03:21:19.503477Z","shell.execute_reply":"2023-06-05T03:21:20.688904Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# vgg19_model.load_weights(\"/kaggle/input/vgg19h5/vgg19_process_lung_2.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-05T02:08:41.150747Z","iopub.status.idle":"2023-06-05T02:08:41.15112Z","shell.execute_reply.started":"2023-06-05T02:08:41.150931Z","shell.execute_reply":"2023-06-05T02:08:41.150948Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"vgg19_model.compile(optimizer = tf.keras.optimizers.experimental.Adam(0.0001), loss = 'binary_crossentropy',\n                           metrics = ['binary_accuracy', tf.keras.metrics.AUC()])","metadata":{"execution":{"iopub.status.busy":"2023-06-05T02:08:41.151902Z","iopub.status.idle":"2023-06-05T02:08:41.152316Z","shell.execute_reply.started":"2023-06-05T02:08:41.152072Z","shell.execute_reply":"2023-06-05T02:08:41.152088Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"steps_per_epoch = train_df.shape[0]  // BATCH_SIZE\nsteps_per_epoch","metadata":{"_uuid":"24c2f57a-6370-49c7-8180-f5d68ad1a56c","_cell_guid":"edf7f452-ef07-4a13-a497-bb05e187c9c7","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-06-05T03:21:32.509064Z","iopub.execute_input":"2023-06-05T03:21:32.509423Z","iopub.status.idle":"2023-06-05T03:21:32.515651Z","shell.execute_reply.started":"2023-06-05T03:21:32.509392Z","shell.execute_reply":"2023-06-05T03:21:32.514537Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# vgg19_model.fit(train_data, \n#             steps_per_epoch = steps_per_epoch,\n# #           validation_data = valid_data, \n#             epochs = 1, \n#             verbose = 1,\n#           )","metadata":{"_uuid":"6b4ae753-0749-4bdb-8d8d-8fddbeed6ba4","_cell_guid":"758fdf21-3c69-4c01-966a-b74cf3ab1a4c","collapsed":false,"jupyter":{"outputs_hidden":false},"execution":{"iopub.status.busy":"2023-06-05T02:08:41.158154Z","iopub.status.idle":"2023-06-05T02:08:41.158589Z","shell.execute_reply.started":"2023-06-05T02:08:41.158406Z","shell.execute_reply":"2023-06-05T02:08:41.158424Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### NORMAL","metadata":{}},{"cell_type":"code","source":"vgg19_model.load_weights(\"/kaggle/input/weights/vgg19_no2.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:25.449543Z","iopub.execute_input":"2023-06-05T03:30:25.450219Z","iopub.status.idle":"2023-06-05T03:30:26.755197Z","shell.execute_reply.started":"2023-06-05T03:30:25.450184Z","shell.execute_reply":"2023-06-05T03:30:26.754179Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred_vgg19_normal = vgg19_model.predict(test_X_normal)\ny_pred_vgg19_normal[:5]","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:26.756964Z","iopub.execute_input":"2023-06-05T03:30:26.757302Z","iopub.status.idle":"2023-06-05T03:30:28.502419Z","shell.execute_reply.started":"2023-06-05T03:30:26.757268Z","shell.execute_reply":"2023-06-05T03:30:28.501492Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_vgg19_normal = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_vgg19_normal[:,idx])\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_vgg19_normal += auc(fpr, tpr)\nmean_auc_vgg19_normal /= 14\nprint(mean_auc_vgg19_normal)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:28.504316Z","iopub.execute_input":"2023-06-05T03:30:28.505133Z","iopub.status.idle":"2023-06-05T03:30:29.034458Z","shell.execute_reply.started":"2023-06-05T03:30:28.505096Z","shell.execute_reply":"2023-06-05T03:30:29.033502Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### PREPROCESS","metadata":{}},{"cell_type":"code","source":"vgg19_model.load_weights(\"/kaggle/input/vgg19h5/vgg19.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:32.22431Z","iopub.execute_input":"2023-06-05T03:30:32.234236Z","iopub.status.idle":"2023-06-05T03:30:33.602065Z","shell.execute_reply.started":"2023-06-05T03:30:32.234188Z","shell.execute_reply":"2023-06-05T03:30:33.600982Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred_vgg19_process = vgg19_model.predict(test_X_process)\ny_pred_vgg19_process[:5]","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:39:37.013555Z","iopub.execute_input":"2023-06-05T03:39:37.013956Z","iopub.status.idle":"2023-06-05T03:39:38.339431Z","shell.execute_reply.started":"2023-06-05T03:39:37.013922Z","shell.execute_reply":"2023-06-05T03:39:38.33843Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc,f1_score\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_vgg19_process = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_vgg19_process[:,idx])\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_vgg19_process += auc(fpr, tpr)\n\nmean_auc_vgg19_process /= 14\nprint(mean_auc_vgg19_process)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:40:13.295972Z","iopub.execute_input":"2023-06-05T03:40:13.296342Z","iopub.status.idle":"2023-06-05T03:40:13.776396Z","shell.execute_reply.started":"2023-06-05T03:40:13.29631Z","shell.execute_reply":"2023-06-05T03:40:13.775518Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## EfficientNetV2S ","metadata":{}},{"cell_type":"code","source":"input = tf.keras.layers.Input([IMAGE_SIZE, IMAGE_SIZE, 3], dtype = tf.float32)\n\neff = tf.keras.applications.EfficientNetV2S(include_top = False,weights = 'imagenet') \neff = eff(input)\n\nnormal = BatchNormalization() (eff)\nflatten = GlobalAveragePooling2D() (normal)\n\ndense1 = Dense(256,activation = 'relu') (flatten)\n\noutput = Dense(len(all_labels), activation = 'sigmoid') (dense1)\n\neff_model = tf.keras.Model(inputs=[input], outputs=[output])\neff_model.summary()","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:40.70704Z","iopub.execute_input":"2023-06-05T03:30:40.707407Z","iopub.status.idle":"2023-06-05T03:30:48.406821Z","shell.execute_reply.started":"2023-06-05T03:30:40.707376Z","shell.execute_reply":"2023-06-05T03:30:48.40577Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"eff_model.compile(optimizer = tf.keras.optimizers.experimental.AdamW(0.0001), loss = 'binary_crossentropy',\n                           metrics = ['binary_accuracy', tf.keras.metrics.AUC()])\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:48.408756Z","iopub.execute_input":"2023-06-05T03:30:48.40918Z","iopub.status.idle":"2023-06-05T03:30:48.44618Z","shell.execute_reply.started":"2023-06-05T03:30:48.409145Z","shell.execute_reply":"2023-06-05T03:30:48.445312Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"steps_per_epoch = train_df.shape[0]  // BATCH_SIZE\nsteps_per_epoch\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:50.570805Z","iopub.execute_input":"2023-06-05T03:30:50.571351Z","iopub.status.idle":"2023-06-05T03:30:50.583786Z","shell.execute_reply.started":"2023-06-05T03:30:50.571309Z","shell.execute_reply":"2023-06-05T03:30:50.582655Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# eff_model.fit(train_data, \n#             steps_per_epoch = steps_per_epoch,\n# #           validation_data = valid_data, \n#             epochs = 1, \n#             verbose = 1,\n#           )\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:30:53.767053Z","iopub.execute_input":"2023-06-05T03:30:53.767415Z","iopub.status.idle":"2023-06-05T03:30:53.77435Z","shell.execute_reply.started":"2023-06-05T03:30:53.767384Z","shell.execute_reply":"2023-06-05T03:30:53.773431Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### NORMAL","metadata":{}},{"cell_type":"code","source":"eff_model.load_weights(\"/kaggle/input/weights/effv2s.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:31:36.138412Z","iopub.execute_input":"2023-06-05T03:31:36.139607Z","iopub.status.idle":"2023-06-05T03:31:38.143902Z","shell.execute_reply.started":"2023-06-05T03:31:36.139535Z","shell.execute_reply":"2023-06-05T03:31:38.142733Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred_eff_normal = eff_model.predict(test_X_normal)\ny_pred_eff_normal[:5]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:31:38.148004Z","iopub.execute_input":"2023-06-05T03:31:38.150146Z","iopub.status.idle":"2023-06-05T03:31:40.336069Z","shell.execute_reply.started":"2023-06-05T03:31:38.150112Z","shell.execute_reply":"2023-06-05T03:31:40.334609Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_eff_normal = 0\nf1_eff_normal = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_eff_normal[:,idx])\n    f1_eff_normal += f1_score\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_eff_normal += auc(fpr, tpr)\n\nmean_auc_eff_normal /=14\n\nprint(mean_auc_eff_normal)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:31:46.75064Z","iopub.execute_input":"2023-06-05T03:31:46.751186Z","iopub.status.idle":"2023-06-05T03:31:47.254114Z","shell.execute_reply.started":"2023-06-05T03:31:46.751145Z","shell.execute_reply":"2023-06-05T03:31:47.253219Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### PREPROCESS","metadata":{}},{"cell_type":"code","source":"eff_model.load_weights(\"/kaggle/input/vgg19h5/effv2s_1.h5\")","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:31:55.528792Z","iopub.execute_input":"2023-06-05T03:31:55.529283Z","iopub.status.idle":"2023-06-05T03:31:57.711772Z","shell.execute_reply.started":"2023-06-05T03:31:55.529242Z","shell.execute_reply":"2023-06-05T03:31:57.710776Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred_eff_process = eff_model.predict(test_X_process)\ny_pred_eff_process[:5]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:32:11.674661Z","iopub.execute_input":"2023-06-05T03:32:11.675022Z","iopub.status.idle":"2023-06-05T03:32:14.889583Z","shell.execute_reply.started":"2023-06-05T03:32:11.674993Z","shell.execute_reply":"2023-06-05T03:32:14.88862Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_eff_process = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_eff_process[:,idx])\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_eff_process += auc(fpr, tpr)\n    \nmean_auc_eff_process /=14\nprint(mean_auc_eff_process)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')","metadata":{"execution":{"iopub.status.busy":"2023-06-05T03:32:18.95747Z","iopub.execute_input":"2023-06-05T03:32:18.958342Z","iopub.status.idle":"2023-06-05T03:32:19.567117Z","shell.execute_reply.started":"2023-06-05T03:32:18.958296Z","shell.execute_reply":"2023-06-05T03:32:19.566261Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## ConvNet","metadata":{}},{"cell_type":"code","source":"import tensorflow_hub as hub","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:28:39.358033Z","iopub.execute_input":"2023-06-05T04:28:39.358515Z","iopub.status.idle":"2023-06-05T04:28:39.637459Z","shell.execute_reply.started":"2023-06-05T04:28:39.35848Z","shell.execute_reply":"2023-06-05T04:28:39.636479Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"MODEL_PATH = \"https://tfhub.dev/sayakpaul/convnext_base_21k_1k_224_fe/1\"","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:28:39.714453Z","iopub.execute_input":"2023-06-05T04:28:39.715141Z","iopub.status.idle":"2023-06-05T04:28:39.719025Z","shell.execute_reply.started":"2023-06-05T04:28:39.715077Z","shell.execute_reply":"2023-06-05T04:28:39.71813Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"input = tf.keras.layers.Input([IMAGE_SIZE, IMAGE_SIZE, 3], dtype = tf.float32, name = 'input')\ninput_conv = tf.keras.layers.Resizing(224,224, name = 'resize_input') (input)\n\nhub_layer = hub.KerasLayer(MODEL_PATH, trainable=True, name = \"ConvNet\")\nconv_model = hub_layer (input_conv)\n\ndense1 = Dense(768, activation = tf.keras.activations.gelu, name = \"dense_768\") (conv_model)\ndense1 = Dense(512, activation = tf.keras.activations.gelu, name = \"dense_512\") (dense1)\ndense1 = Dense(128, activation = tf.keras.activations.gelu, name = \"dense_128\") (dense1)\noutput = Dense(len(all_labels), activation = 'sigmoid',name = 'output') (dense1)\n\nconv_model = tf.keras.Model(inputs=[input], outputs=[output])\nconv_model.summary()","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:28:40.279339Z","iopub.execute_input":"2023-06-05T04:28:40.279688Z","iopub.status.idle":"2023-06-05T04:28:53.818903Z","shell.execute_reply.started":"2023-06-05T04:28:40.27966Z","shell.execute_reply":"2023-06-05T04:28:53.817952Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"conv_model.compile(optimizer = tf.keras.optimizers.experimental.AdamW(3e-5), loss = 'binary_crossentropy',\n                           metrics = ['binary_accuracy', tf.keras.metrics.AUC()])\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:28:53.820732Z","iopub.execute_input":"2023-06-05T04:28:53.821839Z","iopub.status.idle":"2023-06-05T04:28:53.848699Z","shell.execute_reply.started":"2023-06-05T04:28:53.821802Z","shell.execute_reply":"2023-06-05T04:28:53.847829Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"steps_per_epoch = train_df.shape[0]  // BATCH_SIZE\nsteps_per_epoch\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:28:53.849837Z","iopub.execute_input":"2023-06-05T04:28:53.850146Z","iopub.status.idle":"2023-06-05T04:28:54.540753Z","shell.execute_reply.started":"2023-06-05T04:28:53.85011Z","shell.execute_reply":"2023-06-05T04:28:54.539781Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# conv_model.fit(train_data, \n#             steps_per_epoch = steps_per_epoch,\n# #           validation_data = valid_data, \n#             epochs = 1, \n#             verbose = 1,\n#           )","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:28:54.543719Z","iopub.execute_input":"2023-06-05T04:28:54.544387Z","iopub.status.idle":"2023-06-05T04:28:54.551529Z","shell.execute_reply.started":"2023-06-05T04:28:54.544352Z","shell.execute_reply":"2023-06-05T04:28:54.550573Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### NORMAL","metadata":{}},{"cell_type":"code","source":"conv_model.load_weights('/kaggle/input/weights/convnet/my_checkpoint')","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:40:08.688593Z","iopub.execute_input":"2023-06-05T04:40:08.688949Z","iopub.status.idle":"2023-06-05T04:40:09.623005Z","shell.execute_reply.started":"2023-06-05T04:40:08.68892Z","shell.execute_reply":"2023-06-05T04:40:09.622113Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred_conv_normal = conv_model.predict(test_X_normal)\ny_pred_conv_normal[:5]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:40:09.624801Z","iopub.execute_input":"2023-06-05T04:40:09.626642Z","iopub.status.idle":"2023-06-05T04:40:14.585002Z","shell.execute_reply.started":"2023-06-05T04:40:09.626607Z","shell.execute_reply":"2023-06-05T04:40:14.584089Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_conv_normal = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_conv_normal[:,idx])\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_conv_normal += auc(fpr, tpr)\n\nmean_auc_conv_normal /= 14\nprint(mean_auc_conv_normal)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:40:14.586501Z","iopub.execute_input":"2023-06-05T04:40:14.586847Z","iopub.status.idle":"2023-06-05T04:40:15.037872Z","shell.execute_reply.started":"2023-06-05T04:40:14.586815Z","shell.execute_reply":"2023-06-05T04:40:15.037037Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### PREPROCESS","metadata":{}},{"cell_type":"code","source":"conv_model.load_weights('/kaggle/input/vgg19h5/conv-checkpoint/my_checkpoint')\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:40:15.040457Z","iopub.execute_input":"2023-06-05T04:40:15.041185Z","iopub.status.idle":"2023-06-05T04:40:16.46803Z","shell.execute_reply.started":"2023-06-05T04:40:15.041148Z","shell.execute_reply":"2023-06-05T04:40:16.467024Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y_pred_conv_process = conv_model.predict(test_X_process)\ny_pred_conv_process[:5]\n","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:40:16.469284Z","iopub.execute_input":"2023-06-05T04:40:16.469632Z","iopub.status.idle":"2023-06-05T04:40:21.423345Z","shell.execute_reply.started":"2023-06-05T04:40:16.469598Z","shell.execute_reply":"2023-06-05T04:40:21.422469Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from sklearn.metrics import roc_curve, auc\nfig, c_ax = plt.subplots(1,1, figsize = (9, 9))\nmean_auc_conv_process = 0\nfor (idx, c_label) in enumerate(all_labels):\n    fpr, tpr, thresholds = roc_curve(test_Y[:,idx].astype(int), y_pred_conv_process[:,idx])\n    c_ax.plot(fpr, tpr, label = '%s (AUC:%0.2f)'  % (c_label, auc(fpr, tpr)))\n    mean_auc_conv_process += auc(fpr, tpr)\n\nmean_auc_conv_process /= 14\nprint(mean_auc_conv_process)\nc_ax.legend()\nc_ax.set_xlabel('False Positive Rate')\nc_ax.set_ylabel('True Positive Rate')\nc_ax.set_title(\"AUC for each class\")","metadata":{"execution":{"iopub.status.busy":"2023-06-05T04:44:45.947189Z","iopub.execute_input":"2023-06-05T04:44:45.947551Z","iopub.status.idle":"2023-06-05T04:44:46.465944Z","shell.execute_reply.started":"2023-06-05T04:44:45.947521Z","shell.execute_reply":"2023-06-05T04:44:46.465079Z"},"trusted":true},"execution_count":null,"outputs":[]}]}